0

Higher Education and Social Mobility: Some Results and Some Questions

Robert B. Archibald

David H. Feldman

Peter McHenry

April 2017

Department of Economics

College of William and Mary

This paper is prepared for the Conference on Higher Education and Social Mobility, April 21 and

22, in Williamsburg Virginia

1

I. Introduction

For generations US identity has been shaped by the proposition that talent and hard

work would allow anyone to rise above the random circumstances of their birth. The US was

not a rigid class-based society, so people could invent or reinvent themselves. Recent data

suggest that this idea of the American Dream is more of an American fantasy. A child born to a

low-income family in the United States has a smaller chance of upward economic mobility than

children born in similar circumstances in many other countries.

Social mobility is commonly measured using the intergenerational elasticity between

fathers’ and sons’ earnings. This elasticity shows how strongly earnings are related across

generations.1 Figure 1 presents Corak’s (2013) estimates of the intergenerational elasticity in

twenty-two nations. In this broad international comparison, fourteen nations exhibit greater

social mobility than the United States.

[Figure 1 Here]

The intergenerational elasticity is quite stable across time in the United States.2 Chetty

et al. (2014, p.1) give a visual analogy: “…envision the income distribution as a ladder with each

percentile representing a different rung. The rungs of the ladder have grown further apart

(inequality has increased), but children’s chances of climbing from lower to higher rungs have

not changed (rank-based mobility has remained stable).”3

1 Ideally, social mobility should be measured in a less gendered way. Data availability has

limited many studies to using fathers and sons. When daughters’ incomes are available the

results are similar. 2 See Tom Hertz (2007), Chul-In Lee and Gary Solon (2009) and Raj Chetty, Nathaniel Hendren,

Patrick Kline, Emmanuel Saez, and Nicholas Turner (2014) 3 Long and Ferrie (2013) show that intergenerational mobility in the US was indeed greater than

in Great Britain in the 19th

century, but that advantage was gone by 1950.

2

A reasonable person might ask, “So what? Why do we need high social mobility?” One

answer is that an economy that allows talent to flourish will be more productive. This is an

efficiency argument. But to think about the optimal amount of social mobility we need a model

of the process that determines social mobility. Gary Solon (2004) models intergenerational

mobility by highlighting the heritability of traits (from nature, nurture, or both), the family’s

human capital investment in their children, and public investment in children’s human capital.4

In Solon’s model social mobility is higher if the correlation between parents’ income and the

traits their children inherit is low, the return to education is low, and public investment in

children is more progressive. The heritability of ability and income-raising behavioral traits is

largely beyond easy policy intervention, and a nurturing family environment is a good thing.

Likewise, a higher return to education isn’t a problem, even though it raises earnings inequality

ceteris paribus. As a result, the optimal intergenerational income elasticity is not zero. But the

capacity of high-income families to invest more in developing their children’s human capital

decreases intergenerational mobility as it reinforces income inequality. On efficiency grounds

alone, a progressive system of public investment in education designed to increase social

mobility is likely to be desirable.

Higher education clearly affects social mobility. First, it affects the productivity of

private human capital investments. Corak’s (2013) data show that there is a strong positive

relationship between the college earning premium and the intergenerational earnings

elasticity. As Autor (2014) suggests, this may seem odd because education should be “the great

equalizer.” But in the United States at least, parents’ education level is a very good predictor of

4 Family bequests also affect mobility, and other models include this effect.

3

a child’s educational attainment, so when the return to education is high children from high-

income families have two big advantages. They get more education, and that education then

pays off with higher earnings over their lifetime.

Progressive public investment in education is needed because the current mechanism

that sorts students into universities still places weight on family income and parental levels of

education independent of the measures of student ability that largely drive academic success.

Some of this weight comes from schools whose need-aware admissions process values family

income. Some comes from families and students who have poor information about the costs

and benefits of various postsecondary options, or whose financial constraints continue to limit

their investment in their children. Social mobility is impeded when high ability students from

high-income families are more likely to go to college than high ability students from low-income

families. Social mobility also is reduced if the institutions high-income students attend receive

more resources. In both cases, income inequality reinforces the effectiveness of private

investment in human capital for the well off. Higher education policy plays a significant role in

determining the progressivity of public investment in education. Pell grants, for instance,

relieve some of the financing constraints on lower-income families. State appropriations for

public colleges and universities are much less targeted because they reduce tuition for all

students.

A substantial body of evidence suggests that the US higher education system as a whole

is not doing the best possible job of enhancing social mobility. Data from the Current

Population Survey show that 77% of dependent students coming from families in the top

quarter on the income distribution earn a bachelor’s degree compared to 9% of students

4

coming from families in the bottom quarter of the income distribution.5 Students who come

from families in the top income quartile do have stronger correlates with college success, but

the correlation between parental income and student academic traits isn’t strong enough on its

own to support these results. For the overall pool of potential college students the cost of

attendance (net tuition plus foregone wages) is high, and financial aid programs are not

sufficient to eliminate the disadvantages of low income. Family income remains a very good

predictor of college completion.

We recognize that the relationship between higher education and social mobility is

complex. In this paper we will highlight two parts of this relationship. First, we will look at the

way our higher education system moves students from high school to college. There are two

questions here. Do the right students go to college, and do students go to the right colleges?

To enhance social mobility we need to get the right students to college and we need to match

students and colleges in the way that takes best advantage of what colleges can offer. Second,

we will look at the distribution of resources among colleges. In order to enhance social mobility

the system of public spending should be progressive. It should funnel funds to colleges that

enroll the students with the most need.

II. Do the Right Students Go to College?

Preparing for college requires students and families to take a set of important sequential

steps. To have a good chance of success, a student needs to take college preparatory courses in

high school, and in many cases this requires making the right curricular choices starting in 8th

grade. Good grades are also important. This requires intelligence, and good behavioral traits

5 These data come from Pell Institute for the Study of Opportunity in Higher Education, (2015).

5

like the capacity to meet deadlines and to do the dull and often repetitive tasks necessary to

achieve longer run goals. Most college-bound students will need to prepare for and take college

entrance examinations. The US higher education system is very forgiving, so if the student has

not taken the right courses, made the right grades, or taken the college entrance exams, there

is still a pathway to a postsecondary credential. Most community colleges and some four-year

colleges are essentially open enrollment institutions. They will take almost any student.6

Despite the fact that there are a large number of open enrollment institutions, the higher

education system should and does reserve most of its spaces for students who demonstrate an

interest in and an aptitude for learning beyond high school.

As the economic payoff to attending college has grown, the number of students who move

directly from high school to college has expanded considerably. In 1980, 49.3% of individuals

were enrolled in a degree granting institution of higher education the year after leaving high

school.7 That percentage increased to 69.2% in 2015. With the expansion of the number of

students going on to college one might expect some diminishing returns to set in. As the

fraction of the high school cohort that goes to college goes up, perhaps the marginal student

entering college would be increasingly poorly prepared. If so, the increased enrollment should

drag down the average preparedness of college students. And many who are worried about the

lack of progress in college completion rates include increased enrollments among the factors

depressing completion rates.

6 In fact most colleges do not even require that students complete high school.

7 These data come from Table 302.2 in the 2016 Digest of Education Statistics.

6

A number of papers have shown that that this concern is largely misplaced.8 Our study

(2015) used two longitudinal data sets (the NLS-72 and ELS:2002) to examine changes in

enrollment behavior between the high school classes of 1972 and 2004.9 Both of these data

sets contain nationally representative samples of high school students who were surveyed

several times after the original survey during their time in high school. In the NLS-72 data 33.4%

of students attended a four-year college during the first three semesters following high school

graduation. The comparable figure from the ELS:2002 data is 47%. This is a substantial (40%)

increase in four-year college attendance. Yet we did not find any decrease in the average

quality of the student body attending 4-year institutions over the thirty-two years between

these two samples. This finding is the result of improved sorting of students into a four-year

college versus the alternatives (a two-year program or starting work directly after high school).

More of the students who were prepared to do well in college in the later sample started

college than in the earlier sample. This is evidence that the US has made some progress at

getting the right students into college.

Figure 2 shows the improved sorting process using the distribution of high school grade

point average (GPA) within each sample.10

We focus on the GPA results because high school

GPA is a better predictor of college success than are the test scores available in the NLS-72 and

ELS:2002. We divided the students in each data set into 10 deciles based on where they fell in

the sample’s GPA distribution and calculated the percentage of students in each decile who

attended a four-year college or university within three semesters of finishing high school. We

8 See Bound, Lovenheim, and Turner (2012) and Archibald, Feldman and McHenry (2015). 9 The ELS:2002 students were sophomores in high school when they were first surveyed.

10 We have information on reading and mathematics test scores, and the same sorting

improvement is apparent using these measures of student ability.

7

calculated GPA deciles within each time period (separately for NLS-72 and ELS:2002), so

changes in grading standards over time (grade inflation) should not influence our results about

college-going behavior. The cross-hatched bars represent the four-year attendance rates for

the NLS-72 sample and the solid bars represent the four-year attendance rates for the ELS:2002

sample. In both surveys the college attendance rate is clearly related to GPA percentile.

Students in the top (10th

) decile have the highest attendance rates, and attendance rates

decline as GPA percentile falls.

[Figure 2 Here ]

In 1972 there was a lot of room to increase college-going rates of students at the top of the

high school class. By 2004 some of that room was filled in. Eighty-one percent of the increase

in four-year attendance rates between the two surveys comes from students in the top half of

the high school GPA distribution. The top three deciles alone generate half of the increase in

four-year college participation, and only 2.2 percent of the rise in participation comes from the

bottom three deciles.

This is the good news. Better-prepared students attended four-year universities in much

greater numbers in 2004 than in 1972. While the number of students attending four-year

colleges increased, the average quality of the students changed very little. Table 1 shows the

average percentile rankings of students in the two surveys for three predictors of college

success, the average GPA percentile and the average percentiles of math and reading tests

given to the two samples. We defined percentiles separately for NLS-72 and ELS:2002 samples,

so overall grade inflation does not explain the tendency of our average preparedness measures

not to diminish.

8

[Table 1 Here]

Diminishing returns must eventually set in. Well-prepared students have non-college

options, and some will choose the armed services, rock stardom, art school, or one of Peter

Thiel’s $100,000 grants to forego college and start a business instead. As a result, 90% might be

the maximum that can be reached even for the highest deciles. Yet there is still quite a bit of

room in the upper deciles, so further increases in the college going rate could still come from

the top of the high school class.

These results suggest that the process of sorting high school seniors into college or work has

improved. This is consistent with students making sound decisions in response to the rising

payoff to a college education. But not all colleges are equal, so there is another kind of match

that is important, the match between students and the particular college they attend. We turn

to this question next.

III. Students And College Matches I

Non-profit colleges and universities in the United States differ from one another on many

dimensions. Public mega-universities with 60,000+ students coexist with small private liberal

arts colleges. Highly selective elite colleges accept less than 5% of their applicant pool while

other schools will take almost everyone who applies. Some colleges graduate 90% of the

students who begin, most within four years. Others struggle to graduate 40% over six years.

Some devote a large amount of resources to education and support services while others

scrape by with much less spending per full time student. How good a job does this complex

system do at promoting social mobility by matching students to the right school?

9

To address this question, we follow a sample of ELS:2002 high school seniors in 2004

who completed high school or attended college in the sample period. We keep only those who

stayed in the sample through the third follow-up (around age 26) so that we can observe

college completion outcomes. We also drop observations with missing values for important

variables.11 Throughout our analysis we apply sample weights so that results are representative

of the population of US high school seniors in 2004 who completed high school.

Table 2 presents evidence about economic segregation and student performance from

the 3rd follow-up of the Education Longitudinal Study of 2002. We break the sample into

quartiles of the socioeconomic status (SES) distribution and by the Barron’s selectivity category

of the college attended for students who start college within two years of completing high

school. To avoid small sample sizes in some of the cells of the table, we combined the top two

and the bottom two Barron’s categories.12

11

Our data set includes restricted-access Barron’s selectivity categories, so we round all sample

sizes to the nearest 10 for confidentiality. Our sample of high school completers who continued

to the 3rd follow-up includes 9,380 respondents. We drop 420 respondents for missing SES

quartile. We drop 350 respondents who attended a four-year college whose Barron’s selectivity

category was missing. We drop 30 respondents whose Barron’s selectivity category was

“Special.” Our final sample size is 8,570. 12

There are two additional data issues we need to note. First, in column (1) there are different

numbers of students in each SES quartile. We are using the ELS:2002 variable BYSES1QU, which

categorizes sample members at tenth grade based on father’s education, mother’s education,

family income, father’s occupation, and mother’s occupation. The original data have roughly

equal numbers of respondents in the bottom three SES quartiles and slightly more in the

highest. Differential high school dropout will tend to increase our representation of highest-

quartile students, since we select only those tenth-graders who were twelfth-graders two years

later. Disproportionate attrition from the sample may also lead to smaller numbers of

observations in the lower SES quartiles. Our use of sample weights should correct for at least

some of that attrition. Second, in column (7), labeled “Did not attend,” we list a 2% college

graduation rate for the entire sample (northeast most block). Graduation usually requires

attendance! We define “did not attend” as not moving into college within the first two years of

10

[Table 2 Here]

The columns tell a clear story about socio-economic segregation. Column (2) shows us that

940 students (roughly 11% of the sample) attended an institution that was categorized as most

or highly competitive by Barron’s. Of these 940 students 660 or 70.25% came from families in

the highest SES quartile, and only 40 or 4.3% came from families in the lowest SES quartile. In

the latest data from the ELS, and in earlier data from other longitudinal data sets, students from

the highest SES quartile predominate at the most selective institutions, and the lowest SES

quartile students make up a very small fraction of the student body at these schools.13

The

dominance of the highest SES quartile students decreases as the selectivity rating of the college

declines, but even at the colleges labeled “competitive” the highest SES students have the

largest share of the student body (40.7%).

The rows are revealing as well. For students from the lowest SES families, 69.1% do not

attend a four-year college; 38.9% (630/1,620) did not attend any college at all in the first two

years after high school and 30.2% (490/1,620) attended a two-year program. For students from

the highest SES families only 21.6% did not move into a four-year program; 5.8% (170/2,920)

did not attend college at all and 15.8% (460/2,920) attended a two-year college. High SES

students are more likely to move directly into a four-year program than low SES students, and

they attend more selective institutions.

finishing high school. Some students in our “did not attend” category did in fact enter college

more than two years after completing high school. Some finished a degree. 13 Bastedo and Jaquette (2011) did a similar calculation for the ELS, and for three other well-

used longitudinal data sets (NLS-72, High School and Beyond in 1980 and NELS:88). SES

segregation by college selectivity is not a new phenomenon, and our conclusion about SES

segregation in enrollment patterns in the ELS data isn’t what is new in Table 2.

11

The ELS data also permit us to examine the relationship between graduation rates and

college selectivity. In each cell we list the actual average graduation rate performance of the

students in that cell and the average predicted graduation rate for that group of students. The

predicted graduation rate is based on a regression that includes the students’ GPA percentile

and their percentiles on tests of mathematics and reading.14 Graduation performance is

strongly related to selectivity. Students who attend highly selective institutions graduate at

higher rates than students who attend less selective institutions. Not surprisingly, much of this

correlation is driven by average readiness differences (GPA and test scores) across selectivity

categories and income quartiles. But we also see that more selective colleges and universities

“produce” graduation rates above predicted values for students in all income quartiles, while

non-selective and two-year programs do the opposite.

The first column demonstrates that college graduation and SES are clearly related in the

overall sample. As we move down the column, the average actual graduation rate diminishes,

from 65.3% for the highest SES category to 19.5% for the lowest SES category. The predicted

graduation rate also diminishes but by a much smaller amount, from 54.7% for the highest SES

category to 25.9% for the lowest SES category. The highest SES category students outperform

the prediction by 10.6% (65.3%-54.7%). This is likely the result of several factors. First, higher

SES students are more likely to attend four-year colleges, and they are concentrated in the

14

The prediction comes from a probit model in which four-year college (bachelor’s level)

completion is the dependent variable. The independent variables are students’ percentiles

within the sample-wide distributions of three measures of college preparedness: high school

GPA and two aptitude tests taken in the tenth grade (in mathematics and reading). The pseudo

R-squared of the probit estimation is 0.2586. The marginal effects (and standard errors) of GPA,

math score, and reading score percentiles are 0.006 (0.0002), 0.0026 (0.0003), and 0.001

(0.0003) respectively.

12

more selective, higher resourced colleges. Second, they are much less likely to drop out of

college for economic reasons than are students from low SES families. The reverse is true for

students from the lowest SES families. These students underperform the prediction by 6.4%

(19.5%-25.9%). The lowest SES students are less likely to attend four-year schools, go to less

selective and less well-resourced institutions when they do attend, and are more likely to drop

out of college for economic reasons.15

In column (2) we see that the disparity in actual graduation rates evident in the full sample

is absent for the students who attend colleges in the highest selectivity category. The

graduation rate is quite similar across SES categories, as is the predicted graduation rate. Highly

selective institutions do not seem to be taking any chances with the lower SES students that

they accept. Column (3) demonstrates a similar result. The actual graduation rate is slightly

higher for the highest SES students, but it is roughly equal for the remaining three groups. Only

in column (4) and thereafter do we see a strong and consistent relationship between SES and

the actual graduation rate. Also, except in the highest selectivity category, the colleges appear

to give preferences in admission to students from the lowest SES quartile. This follows because

their predicted graduation rates are below the predicted graduation rates for students from

higher up the SES ladder.

We highlight two main results. Both are variations of the notion that money matters. First,

where a student goes to school is quite important. Going to a more selective, better-resourced

15 Ehrenburg and Webber (2010) have shown that student service expenditures influence graduation

and persistence rates and that the largest effects are at schools that have lower current graduation and

first-year persistence rates. Students from low SES backgrounds disproportionately attend these under-

resourced institutions.

13

school leads to better outcomes.16 The actual graduation rates decline as school selectivity rank

falls. This is true for the full sample and for every SES group. Second, the resources the family

can provide also matter. With only a few exceptions the actual graduation rate declines as the

students’ SES moves down the ladder. In addition, the highest SES group is the only one for

which the actual graduation rate outperforms the predicted graduation rate for all categories,

including two-year community colleges.

IV. Students and College Matches II

The data in table 2 are based on the actual matches between colleges and students.

These matches reflect how well students navigate the application process and how colleges

react to the applications they see. Socio-economic status clearly affects both parts of the

process. Higher SES students have many advantages in preparing college applications. They can

and do take courses to prepare them for admission tests. They are more likely to attend high

schools that send a large number of students to highly selective colleges, so they get good

counseling and have teachers who may have attended selective schools and who are used to

writing good letters of recommendation. And they can afford to apply to the recommended

number of reach, match, and safety schools. Also, higher SES students are very appealing to

tuition-dependent colleges. Some high SES applicants are legacies who receive preferences in

admission. In the regular admission process at need-aware schools high-income applicants may

receive preference at the margin, and the margin may be very wide. And when a college is

16

Caroline Hoxby (2016) measures the productivity of additional years of schooling across

colleges with different average selectivity (SAT) profiles. She shows that productivity, measured

as value added by the school relative to the total social investment in its students is roughly

constant across schools whose admissions are at least minimally selective. In that case, moving

resources toward less selective schools is just as productive as additional spending at more

selective schools while improving social mobility.

14

deciding which students to take from the wait list, a student with a lower financial aid

requirement has an advantage. To study the effects of SES on the application and admission

process we need to create a hypothetical match between students and colleges that ignores

SES status to compare with the actual matches.

We will use a simple example to illustrate our approach. Assume there are two colleges

both with a capacity of 10 students. On some objective measures College A ranks above

College B. Further assume there are 20 students, and we can rank them from 1 to 20 on some

scale of academic promise. Given these assumptions a perfect matching would place students

ranked 1 to 10 in College A and students ranked 11 to 20 in College B.17

There are two cases to consider. First, assume that College A has a strong desire to have

a full French horn brigade, and to do so it has to recruit two good French horn players in this

year’s class. Also, assume that the best French horn players are students ranked 15 and 16.

Giving spots to these two students results in College A admitting students ranked 1 to 8, 15,

and 16, and College B having a class of students ranked 9 to 15 and 17 to 20. As a result of

College A’s desire for French horn players, the two French horn players are over-matched. They

are attending a more highly ranked school than they would if admission were solely based on

academic promise. Also, students ranked 9 and 10 are under-matched.18 They are attending a

17

See Sallee, Resch, and Courant (2008) for a rationale for matching the best students with the

best schools. 18

This kind of under-matching has been the source of much of the legal wrangling about racial

preferences in college admission. In some instances students have initiated a legal battle,

arguing that a member of a minority group who received a preference in admission took a spot

they “deserved.” Some examples that reached the Supreme Court include Regents of the

University of California v. Bakke, which was decided in 1978, and Gratz v. Bollinger and Grutter

v. Bollinger, both of which were decided in 2003.

15

less highly ranked school than they would if admission were solely based on their academic

promise.

The second case flips the causation. Suppose that the students ranked 5th and 6th are

unaware that College A would have accepted them, and they only apply to College B. When the

students sort out in this case, College A would admit students 1 to 4, and 7 to 12, and College B

would admit students 5, 6, and 13 through 20. In this case, students 5 and 6 are under-

matched. They are attending a less highly ranked school than they would attend if admission

were solely based on academic promise. And students 11 and 12 are over-matched. They

gained admission to a more highly ranked school than they would have attended if admission

were solely based on academic promise.

The two cases illustrate the adding up constraints inherent in this approach. In the first

case, a given amount of over-matching creates an equal amount of under-matching. In the

second case, a given amount of under-matching creates an equal amount of over-matching.

Given the wide set of admission preferences used by colleges and the idiosyncratic choices

made by students when they select where to go, under-matching caused by over-matching and

over-matching caused by under-matching is a common occurrence.

Our simple example suggests some additional terminology. We can identify two kinds of

mismatching, primary mismatching and secondary mismatching. The over-matching of the

French horn players in the first case and the under-matching of the students who did not apply

to the best school they could have been admitted to in the second case are examples of

primary mismatching. The secondary mismatching is the bumping of students 9 and 10 from

College A to College B in the first case and the elevation of students 11 and 12 from College B to

16

College A in the second case. With more than two selectivity categories, secondary

mismatching can only involve a one-level college mismatch. But primary mismatching can

involve both one-level college mismatches and multi-level college mismatches. If we observe a

two or more level mismatch, that must be a primary mismatch. The student or the university

made a choice to do something, or not to do something. A one level mismatch could be primary

or secondary.

We can learn about primary mismatching in the third follow-up of the ELS:2002 by

creating an academic promise score for each student. The score is a weighted average of the

student’s high school GPA percentile and the student’s percentiles on the math and reading

test scores. The weights were the coefficients from our regression predicting college

graduation.19 We then created a hypothetical assignment of students to institutions ranked by

Barron’s selectivity categories. For example, in the data in Table 2 the colleges rated most or

highly selective by Barron’s actually enrolled 940 students from our sample. Our hypothetical

assignment put the students with the 940 highest academic promise scores in this group of

schools. The next group of students, those ranked 941 to 2211, was put into the 1,270 slots in

colleges rated very competitive in Barron’s, and the other groups of colleges were filled in a

similar fashion.

Table 3 presents the results of this exercise. To understand the table compare column

(2) in table 2 with column (1) in table 3. Column (2) in table 2 shows that 940 students in the

sample attended a college rated most or highly selective by Barron’s. This column also gives the

19

The academic promise score is the prediction of the likelihood of four-year college graduation

conditional on high school GPA and math and reading aptitude test scores from the probit

model described earlier.

17

breakdown of these students by SES quartiles. Column (1) in Table 3 shows the result of

allocating 940 students to most and highly selective schools based on their academic promise

scores. Column (1) in Table 3 shows that this allocation mechanism fills the 940 slots in most

and highly selective colleges with 560 students from the highest SES quartile, 220 from the third

SES quartile, 110 from the second SES quartile, and 50 from the lowest SES quartile. Column (2)

in Table 2 shows that the actual distribution of students to these institutions had 660 students

from the highest SES quartile, 170 from the third SES quartile, 70 from the second SES quartile,

and 40 from the lowest SES quartile. Reallocating the students takes 100 slots from students in

the highest SES quartile in the actual allocation and moves 50 (220-170) of them to students in

the third SES quartile, 40 (110-70) to students in the second SES quartile, and 10 to students in

the lowest SES quartile.

[Table 3 Here]

Using academic promise scores to reallocate students does not eliminate the

relationship between SES and attendance at elite institutions. A large percentage of the

students with the top grades and scores do come from the top of the SES ladder. Yet the

hypothetical allocation is less heavily weighted toward high SES students. The other entries in

the table allow us to investigate more fully the differences between the actual and hypothetical

allocation of students.

The entries in columns (2) through (7) allow us to understand mismatching in the ELS

data. The cells in each row give the percentage of the total number of students our algorithm

has assigned to that particular selectivity level that actually went to each type of school (or into

the workforce). For example, the first row of the table tells us that 47.2 percent of the students

18

in the highest SES quartile who our hypothetical assignment put in the most or highly

competitive group of colleges actually attended a college in that group, while 27.7 percent of

these students attended a college in the very competitive group. The percentages in each row

must add to 100. Compared to the other entries in the table, the 47.2 percent of students

actually attending a college consistent with their hypothetical placement is quite high. It

represents a “correct” match. The entries down the principle diagonal represent these correct

matches, and the percentages are usually high, though seldom as high as 47.2 percent.

The highest percentage of correct matches in the table is the lower right. These are the

lowest SES students who are not assigned to any college by our scoring procedure. Almost two

thirds of these lowest SES students know that they are not college material, so they do not even

try college.20 This compares with 36.4 percent of similarly under-qualified highest SES quartile

students. Unlike their lower SES colleagues, underprepared high SES students are more likely to

give college a try than not.

The off diagonal elements represent incorrect matches. These students attended an

institution inconsistent with their academic promise scores. As we argued above, a one

competitiveness category mismatch will include all of the secondary mismatches as well as

some primary mismatches. The two or more category mismatches will all be primary

mismatches, and these mismatches will be the focus of our analysis.

The entries above the diagonal represent under-matches, where a student goes to a

college ranked below the level predicted by our hypothetical assignment. For example, the first

20 When we speak of “correct” matches, that term is based on 12th grade preparedness. The

mobility gains from progressive higher education policy underestimate the potential gains from

interventions that might affect the curricular choices and behavior of lower SES students at a

much earlier age.

19

row of column (5) shows that 2.3 percent of the highest SES quartile students that our

hypothetical assignment would have placed in a most or highly competitive college actually

attended a college ranked less- or non-competitive. Using the data from table 2, these students

attended a college with a graduation rate of 24.1 percent when their academic qualifications

suggest they could have been admitted to a school with an over 80 percent graduation rate.

These under-matched students are not taking advantage of opportunities that would have been

available. Looking at the situations in which students are under-matched by at least two

selectivity categories (the last four columns for the students hypothetically assigned to the

most and highly selective colleges, the last three columns for students hypothetically assigned

to a very selective college, the last two columns for students hypothetically assigned to a

competitive college, and the last column for students hypothetically assigned to a less or non-

competitive college) the data indicate that primary under-matching is clearly related to SES. For

example, consider the students hypothetically matched with a very competitive college who

actually attended a two-year college. This group includes 6.7 percent of students in the highest

SES quartile, 15.5 percent of students in the third SES quartile, 18.5 percent of students in the

second SES quartile, and 23.5 percent of students in the lowest SES quartile. The lower the SES

the more prevalent is under-matching.

Students are over-matched in the cells below the diagonal. We find the reverse result

for the relationship between SES and mismatching in the case of primary over-matching, which

occurs where students are over-matched by two or more college competitiveness groups. For

the most or highly competitive category (column 2) this includes the last four rows. For the

highly selective category (column 3) this includes the last three rows. For the competitive

20

category (column 4) this includes the last two rows, and for the less or non-competitive

category (column 5) this includes the last row. Primary over-matching is more prevalent among

high SES students. For example, consider the students hypothetically placed in a two-year

college who actually attended a very competitive college. This group includes 10.9 percent of

students in the highest SES quartile, 4.2 percent of students in the third SES quartile, 2.1

percent of the students in the second SES quartile, and 2.6 percent of the students in the

lowest SES quartile.21

Table 4 presents proportions of various groups who are under-matched and over-

matched in the ELS sample.22 We include entries for two or more category mismatches and

three or more category mismatches. In each case we give the base rate, which is the proportion

of all students with the possibility of being under- or over-matched by the requisite number of

categories. The sample size for the three or more category mismatches is 4,110, and the sample

size for the two or more category mismatches is 4,750.23

There are some very clear patterns in the results. First consider under-matches. Gender

and race do not appear to be strongly related to the likelihood of being under-matched.

21

Our result is similar to what Dillon and Smith (2013) find using an earlier and smaller data set

(the NLSY97). They find that under-matched students are more prevalent in the lower quartiles

of the family wealth distribution and over-matched students are more prevalent in the upper

quartiles of the family wealth distribution. 22

Sample weights were used for Table 4. Base rate refers to the average in the sample of

respondents who could possibly have been over-matched or under-matched. SES quartiles are

calculated based on 10th grade family income, mother's completed education and occupational

prestige, and father's completed education and occupational prestige. 23

The population for base rates of two or more category under-matches includes students who

were hypothetically matched to colleges with Barron’s ratings of most or highly competitive,

very competitive, competitive, and less or non-competitive. This group has a population of

4,750 students. The population for the base rates of three or more category under matches

excludes the 640 students at colleges rated by Barron’s as less or non-competitive.

21

Students who are black or Asian are slightly less likely to be under-matched, and Hispanics are

slightly more likely to under-match, but the differences with the base rates are small. The

results for SES replicate the findings in the previous table. The further up the SES scale is a

student’s family the less likely he or she is to under-match. The same holds for parents’

education. Students from the Northeast are less likely to under-match, and students from the

West are more likely to under-match. Students from rural high schools are also more likely to

under-match. College athletes are less likely to be under-matched, but this result probably

overstates the importance of athletic participation. This happens because all of the athletes

were in college, so being under-matched into the “did not attend” category was not possible for

them. Finally, our proxy for the quality of the student’s high school curriculum (the Advanced

Placement class variable) is not associated with under-matching.

Over-matches are more common for students who are male or black, but not more

prevalent for Hispanics. Over-matches are also related to family SES. The two lowest SES

quartiles are least likely to be over-matched, while the highest SES quartile is more likely to be

over-matched. This is the pattern one would expect to see if a significant number of schools

practice need-aware admissions. In these cases, institutions admit students without regard to

financial need until they have exhausted the funds they have committed to financial aid and fill

the rest of their class with students who do not require financial aid. This will cause them to

admit some high SES students who are not as well qualified as some lower SES students whom

they fail to admit. Parental education shows a similar pattern. The more highly educated are

parents the more likely are their children to be over-matched. Both the results for SES and

parental education may reflect the use of legacy advantages in admission. If schools give special

22

consideration to legacies this could lead to some over-matching, though a 3 or more category

over-match would suggest a very strong legacy preference. The regional pattern reverses the

under-matching results. Students from the Northeast, who were less likely to be under-

matched, are more likely to be over-matched, and students from the West, who were more

likely to be under-matched, are more likely to be over-matched. The same pattern of results

holds for students from rural high schools. Over-matches are very prevalent among college

athletes, particularly over-matches involving 3 or more categories. This is not surprising since

coaches recruit almost all athletes, even those not receiving scholarships. Also, athletic ability

and academic ability are unlikely to be strongly positively correlated. Again, our AP class

variable does not have an effect.

Over-matching and under-matching have generated a significant literature, much of

which is parallel or non-overlapping. Some of the literature on over-matching explores how

colleges try to recruit diverse classes of students. Acting affirmatively could create primary

overmatches if institutions enroll students who are not prepared for the rigors of a highly

selective institution.24 In our data we see little evidence of primary over-matching of students

from low SES backgrounds. Schools also may admit students on the basis of talents that are not

strongly correlated with the measures (GPA and test scores) that we use as predictors of

academic success. Athletic and artistic talents are good examples. We have already noted that

primary over-matching seems more acute among high SES students, and this could reflect

24

Much of the over-match literature focuses on affirmative action policies, particularly in law

schools. See Richard Sander (2004) and Peter Arcidiacono and Michael Lovenheim (2016).

23

legacy preferences, need-aware admission practices, and recruiting athletes from higher SES

backgrounds.25

Much of the concern about under-matching stems from evidence that many high-

performing low-income students have very poor information and do not apply to the colleges

that are a better fit for them academically, despite the fact that these institutions might well

offer enough financial aid to make the higher ranked college less expensive than the institutions

the students actually attend.26 The Expanding College Opportunity (ECO) project (see Hoxby

and Turner, 2013) shows how an inexpensive intervention that provides low-income students

with important and personalized information about their options can have a dramatic impact.

The ECO project gave high-achieving low-income students accurate information about net price

at elite schools for which they were qualified and helped them to understand the application

process. The treated group applied to more schools, and to better schools that offered more

resources. Significantly more students from the treated group were accepted to, and ultimately

enrolled at, the schools for which they were better matched academically. These results show

that higher education markets are far from frictionless. There are many barriers, real and

perceived, that thwart good decision-making, particularly for low SES students and students

who are the first in their family to attend college.

V. Progressivity of Funding

25

Shulman and Bowen (2001) show that affirmative action toward athletes was much more

pronounced than affirmative action toward minorities. See Figure 2.3 page 41) 26

See Hoxby and Avery (2013). Some of the earlier work on under-matching focused on

samples from limited areas. For example Bowen, Chingos and McPherson (2009) investigated a

sample of North Carolina students, and Roderick, Nagaoka, and Allensworth (2006) studied

students from Chicago public schools. More recently nationally representative samples have

also been used; see Dillon and Smith (2013) and Smith, Pender, and Howell (2013).

24

To emphasize a point that should have already come through loud and clear, Table 5

rearranges the data in Table 2 to highlight an important message. College attendance remains

highly related to socioeconomic status. These data give the percentage of each SES quartile in

the ELS sample that attends college by the Barron’s selectivity categories. For example, only

2.5% of students from the lowest SES quartile attend a college in the top Barron’s category

(most or highly competitive). This compares to 22.5% of students in the highest SES category

who attend a college in the top Barron’s category. Students from the bottom half of the SES

ladder do not attend 4-year colleges as frequently as students from the top half of the SES

ladder. They are less likely to attend colleges in the top Barron’s selectivity categories than are

students from the top of the SES ladder. And they are more likely to move directly into the

labor force.

[Table 5 Here]

To enhance social mobility, public funding of higher education needs to be progressive.

Funding should increase as socioeconomic status declines. Effective progressive funding

programs need to shift resources to where the lowest SES students attend college. Some public

funding clearly is progressive. The Pell Grant program is need based. Only relatively low-income

students qualify for a Pell Grant, and among low-income students the grant award grows as

income decreases. The result is clearly a progressive transfer.

State need-based financial aid programs are also progressive transfers. The shift away

from need-based financial aid programs to merit-based ones like Georgia’s HOPE scholarship

program has reduced the progressivity of state financial aid. The data in Tables 3 and 4 clearly

25

show that students higher up the SES ladder have higher high school grades and test scores, so

basing aid on these criteria will lead to a regressive transfer.27

Other funding is not as progressive. The largest single source of support for students

comes from the state subsidy given to public institutions. This subsidy allows state supported

institutions to provide an education at a discount to state residents. Table 6 presents data from

the Department of Education’s Integrated Postsecondary Education System (IPEDS). The data

are in real dollars (deflated by the Consumer Price Index) and averages are weighted using full-

time equivalent attendance weights. The data are for two three-year time periods, 1988-1990

and 2013-2015. The three-year averages smooth out any unusual data for a particular year. The

table reports state appropriations, expenditures on instruction, and expenditures on other core

activities (student services, academic support, and instructional support) for 2013-2015, for

1988-90, and for the difference between those years.

[Table 6 Here]

The real value of state appropriations per full-time equivalent student is not distributed

progressively. State appropriations per student are much higher at the top Barron’s selectivity

categories where more affluent students are more likely to attend college and lower at the less

selective Barron’s categories where less affluent students are more likely to go. Real state

appropriations per full time student have fallen over this time period, and the evidence in Table

6 suggests that the more selective institutions have experienced larger reductions in state

27

One argument in favor of programs like the HOPE scholarship is that they are simple to

understand. As a result, they may send incentives that get more students to take steps earlier in

their schooling to prepare for college.

26

appropriations than the less selective institutions. In a sense, state cuts have been progressive.

This might have been the result of each category of institution taking the same percentage

reduction, but it was not. The most elite category (most and highly competitive) experienced a

39.6 % reduction, the very competitive category experienced a 37.9 % reduction, the

competitive category experienced a 36.0 % reduction, the non- or less competitive category

experienced a 25.3 % reduction, and the 2-year colleges experienced a 20.7 % reduction. In the

aggregate, states protected their less competitive institutions from the worst of the cuts. Still,

despite this protection, the regressive pattern persists. The most recent data on state

appropriations indicate that the more competitive a college is the more it receives in state

appropriations per student.

The data on expenditures show a similar pattern. The more competitive the institution is

the larger are its expenditures on instruction and other core activities. This follows because the

more competitive institutions receive larger state appropriation and also because they are able

to generate more non-state funds. The difference in state appropriations between the most

and highly competitive colleges and the very competitive colleges is $2,014, while the

difference in two spending categories (instructional and other core) combined is $8,287. The

only way this is possible is if the most and highly competitive institutions were able to generate

much more in tuition, grants and contracts, sales of educational services, and/or private funds.

The fact that the most selective public institutions increasingly can generate their own revenue

enhances the regressive nature of the public system as a whole. Both state appropriated funds

and campus generated funds are larger at the institutions that tend to have the most affluent

student bodies.

27

As states decrease the per-student funding of public institutions the finances of these

institutions are increasingly being privatized. Raising funds from all of these non-state sources is

easier for more highly rated public institutions than for schools with lower rankings. This

problem is more acute in states where demographic change is reducing the size of graduating

high school classes, as it is in the Northeast and Midwest.28 In these states, less selective state

universities are likely to lose students (and the associated tuition revenue) to their more highly

rated state competitors. These less-selective state institutions are more frequently used as

backup choices for students who would prefer to attend a state flagship. Also, less selective

institutions are not as effective at generating private giving. The public colleges with lower

ratings are typically newer and often smaller than their more highly rated state-supported

brethren. This means they have a smaller alumni base and fewer chances to land a really big gift

that can make a substantial difference. Finally, most public support for research grants and

contracts is given to institutions with large graduate programs. In most cases (though not all),

these research universities have fairly competitive undergraduate admissions, so the extra

research funding provides benefits that are not readily available to lower ranked public

institutions.29

To conclude, some public support is given on the basis of need, and increasing this type

of public support for higher education will enhance the progressivity of public support and

improve social mobility. The vast majority of federal support for students authorized by Title IX

of the Higher Education Act is based on need, or at least has income cutoffs for eligibility, so it is

28

See Prescott and Bransberger (2012). 29

For a fuller discussion of how changes in the economic and political landscape likely will

affect flagships versus non-flagships, see chapter 10 of Archibald and Feldman (2017).

28

progressive. Need-based financial aid programs at the state level also are progressive. On the

other hand, state support for merit-based financial aid programs and federal support for

research are not progressively distributed. And most importantly state support for public

institutions is distributed regressively. If our objective is to improve social mobility, there is

ample room for changes in public policy.

VI. Summary

We have presented mixed results about the role of higher education in promoting social

mobility. First, we showed that the process of sorting students into four-year colleges has

improved. As a result, the growth in the fraction of high school students moving directly into

four-year colleges has not reduced the average quality of college students. Other things equal,

this should promote social mobility. Increasing the fraction of the well-prepared students who

attend college emphasizes ability over income as a predictor of college attendance. Unless

income and ability are perfectly correlated this change should improve social mobility.

On the other hand, income is still a very strong predictor of the quality of the college

students attend. Under-matching is overwhelmingly a low SES phenomenon. A large number of

high ability low SES students attend colleges that are under-resourced and have very poor

graduation rates. Similar low SES students who are more appropriately matched attend colleges

with substantially more resources and much higher graduation rates. This prompts the question

of how best to improve college matches for low-income students. The barriers are social,

informational, and economic. Where can we make the most progress? We can move resources

29

to the schools that are currently under-resourced, or we can move students to the right

schools. Or we can do both.

Bibliography

Archibald, Robert B., and David H. Feldman, (2017). The Road Ahead for America’s Colleges &

Universities, Oxford: Oxford University Press.

Archibald, Robert B., David H. Feldman and Peter McHenry (2015). “A Quality-Preserving

Increase in Four-year college Attendance,” Journal of Human Capital, 9 (Fall): 265-297.

Arcidiacono, Peter and Michael Lovenheim (2016). “Affirmative action and the quality-fit trade-

off.” Journal of Economic Literature 54 (1) 3-51.

Autor, David (2014). “Skills, education, and the rise of earnings inequality among the ‘other 99

percent’” Science, 344 (May 23): 843-851.

Bowen, Chingos, and McPherson (2009) Crossing the Finish Line: Completing College at

America’s Public Universities, Princeton: Princeton University Press

Bastedo, Michael and Ozan Jaquette (2011). “Running in Place: Low-Income Students and the

Dynamics of Higher Education Stratification,” Educational Evaluation and Policy Analysis, 33

(September): 318-339.

Chetty, Raj, Nathaniel Hendren, Patrick Kline, Emmanuel Saez, and Nicholas Turner (2014). “Is

the United States Still a Land of Opportunity: Recent Trends in Intergenerational Mobility,”

NBER Working Paper 19844, January.

Corak, Miles (2013). “Inequality from Generation to Generation: The United States in

Comparison,” in R. S. Rycroft. The Economics of Inequality, Poverty, and Discrimination in the

21st

Century (ABC-CLIO): 107-123.

Hertz, Tom (2007). “Trends in the Intergenerational Elasticity of Family Income in the United

States,” Industrial Relations, 46 (January): 22-50.

Hoxby, Caroline M., (2016). The Dramatic Economics of the U.S. Market for Higher Education.

The 2016 Martin Feldstein Lecture. NBER Reporter, Number 3.

30

Hoxby, Caroline M. and Sarah Turner (2013). “Informing Students About Their College Options:

A Proposal for Broadening the Expanding College Opportunities Project,” The Brookings

Institution, http://www.hamiltonproject.org/papers/informing_students_about

_their_college_options

Hoxby, Caroline M., and Christopher Avery (2013). “The Missing ‘One-Offs’; the Hidden Supply

of High-Achieving, Low-Income Students.” Brookings Papers on Economic Activity, (Spring), 1-

65.

Lee, Shul-In and Gary Solon (2009). “Trends in Intergenerational Income Mobility,” Review of

Economics and Statistics, 91 (November): 766-772.

Long, Jason and Joseph Ferrie. (2013). "Intergenerational Occupational Mobility in Great Britain

and the United States since 1850." American Economic Review, 103(4): 1109-37.

Pell Institute for the Student of Opportunity in Higher Education (2015) “Indicators of Higher

Education Equity in the United States,” http://www.pellinstitute.org/downloads/publications-

Indicators_of_Higher_Education_Equity_in_the_US_45_Year_Trend_Report.pdf

Prescott, Brian and Peace Bransberger (2012). Knocking on the College Door: Projections of High

School Graduates, 8th

ed. Bolder Colorado: Western Interstate Commission for Higher

Education.

Roderick, Melissa, Jenny Nagaoka, and Elaine Allensworth (2006). “From High School to the

Future,” Chicago Postsecondary Transition Project.

Sallee, James M., Alexandra M. Resch, and Paul N. Courant (2008). “On the Optimal Allocation

of Students and Resources in a System of Higher Education,” B. E. Journal of Economic Analysis

and Policy 8 (1): 1-24.

Sander, Richard (2004).“A Systematic Analysis of Affirmative action in American Law Schools,”

Stanford Law Review 57 (2): 367-483.

Schulman, James L., and William G. Bowen (2002). The Game of Life: College Sports and

Educational Values, Princeton: Princeton University Press.

Solon, Gary (2004). “A model of intergenerational mobility variation over time and place.” In

Miles Corak (editor). Generational Income Mobility in North America and Europe. Cambridge:

Cambridge University Press.

Webber, Douglas A., and Ronald G. Ehrenberg (2010). “Do Expenditures other than

Instructional Expenditures Affect Graduation and Persistence Rates in American Higher

Education.” Economics of Education Review, 29 (6): 947-958.

31

0.67

0.6

0.58

0.52

0.5

0.5

0.49

0.47

0.46

0.46

0.44

0.41

0.4

0.34

0.32

0.29

0.27

0.26

0.19

0.18

0.17

0.15

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

Peru

China

Brazil

Chile

United Kingdom

Italy

Argentina

United States

Switzerland

Pakistan

Singapore

France

Spain

Japan

Germany

New Zealand

Sweden

Australia

Canada

Finland

Norway

Denmark

Figure 1. Intergenerational Earnings Elasticities

32

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1 2 3 4 5 6 7 8 9 10

Figure 2. 4-year College Attendance Rates by High School GPA

NLS ELS

33

Table 1. Average Percentiles of Students Attending Four-year

Colleges, NLS-72 and ELS:2002

NLS-72 ELS:2002

GPA percentile 64.6 66.2

Math test percentile 65.9 64.9

Reading test percentile 63.3 63.8

0

Table 2. Actual and Predicted Graduation Rates by College Attendance (College Type and Selectivity), ELS:2002

Four-year college completion by 3rd follow-up of ELS:2002

(1) (2) (3) (4) (5) (6) (7)

Category of first post-secondary institution attended

All

Most or Highly

competitive (4-

yr)

Very

competitive (4-

yr)

Competitive (4-

yr)

Less- or non-

competitive (4-

yr)

Two-year

college Did not attend

All SES groups

Actual rate 40.5 82.9 79.8 63.8 42.6 24.1 2

Predicted rate 40.5 72.9 64.6 51.9 37.4 30.8 18.1

Observations 8,570 940 1,270 1,890 640 2,170 1,670

Highest SES quartile

Actual rate 65.3 84.5 84.2 72.9 54.8 36.8 8.5

Predicted rate 54.7 73.9 65.9 55.6 45.4 34.6 22.8

Observations 2,920 660 700 770 170 460 170

Third SES quartile

Actual rate 41.9 83.5 75.2 60 46.1 27.8 2

Predicted rate 42.5 71.3 64.6 51.9 37.7 35.8 18.5

Observations 2,160 170 320 560 170 570 370

Second SES quartile

Actual rate 27.5 70.7 73.2 59.1 37.6 18.9 1.9

Predicted rate 33.9 70.7 63.8 49.8 36.2 28.4 20.1

Observations 1,870 70 150 350 150 650 500

Lowest SES quartile

Actual rate 19.5 83.3 76.7 54.4 31.1 17 .5

Predicted rate 25.9 70.6 58.2 44.2 30.3 25.4 15

Observations 1,620 40 100 210 140 490 630

NOTES: Data from ELS:2002. Sample of 2004 seniors. Sample sizes rounded to the nearest 10 for confidentiality. Sample weights used. SES

quartiles calculated based on 10th grade family income, mother's completed education and occupational prestige, and father's completed

education and occupational prestige. Predicted graduation rate based on a regression of four-year college completion (indicator) on high school

GPA percentile, math and reading test score percentiles.

0

Table 3. Comparison of Actual and Counterfactual Post-secondary Attendance, ELS:2002

(1) (2) (3) (4) (5) (6) (7)

Actual college attendance

Merit-based

counterfactual

SES

quartile n

Most or Highly

competitive (4-

yr)

Very competitive

(4-yr)

Competitive (4-

yr)

Less- or non-

competitive (4-

yr) Two-year college Did not attend

Most or Highly competitive (4-yr)

Highest 560 47.2 27.7 19.8 2.3 2.3 .7

Third 220 26.5 25.5 28.3 4.2 14.7 .7

Second 110 27.5 30 22.7 7.5 11.1 1.1

Lowest 50 17.8 32.1 36.1 4.3 9.2 .6

Very competitive (4-yr)

Highest 630 27.3 35.8 26.1 3.4 6.7 .8

Third 330 14.1 27.3 34.4 7.1 15.5 1.6

Second 200 9.4 15.2 41.9 7.4 18.7 7.4

Lowest 110 11.4 20.9 22.6 12.8 23.5 8.8

Competitive (4-yr)

Highest 790 15.6 25.7 33 6.9 15.4 3.5

Third 500 6.3 17.7 36.7 6.7 25.8 6.9

Second 370 2.4 10.1 30.1 8.4 31.8 17.2

Lowest 240 3.2 14.3 26.3 10 30 16.2

Less- or non-competitive (4-yr)

Highest 220 5.1 18 39.3 6.6 23.3 7.7

Third 180 1.7 8.3 38.7 11.5 30 9.7

Second 140 2.6 5.3 16.9 11.5 40 23.8

Lowest 100 0 8.7 25.9 6 38.7 20.6

Two-year college

Highest 510 3.5 10.9 28.5 9.1 31.2 16.7

Third 580 .8 4.2 20.7 10.6 37.2 26.5

Second 570 1.1 2.1 13.1 8.3 43.7 31.6

Lowest 510 .3 2.6 9.5 8.6 39 40.1

Did not attend

Highest 220 .8 .9 7.3 6.6 47.9 36.4

1

Third 350 .9 .3 7.1 8.1 30.4 53.2

Second 490 0 .4 6.4 6.1 36 51

Lowest 610 0 .2 3.4 6 25.7 64.7

NOTES: Data from ELS:2002. Sample of 2004 seniors. Sample weights used. SES quartiles calculated based on 10th grade family income, mother's completed

education and occupational prestige, and father's completed education and occupational prestige. Predicted graduation rate based on a regression of four-

year college completion (indicator) on high school GPA percentile, math and reading test score percentiles.

2

Table 4. Characteristics of Respondents by Mismatch Status, ELS:2002

Average Characteristics of Mismatched Students

(1) (2) (3) (4) (5) (6) (7) (8)

Under-matches Over-matches

2 or more categories 3 or more categories 2 or more categories 3 or more categories

Base rate

Under-

matched Base rate

Under-

matched Base rate

Over-

matched Base rate

Over-

matched

Male 43.1 41.6 41.9 45.7 48.7 53.1 50.5 60.5

Black 4.8 2 3.9 1.3 16.3 24.7 19.8 28.3

Hispanic 7.6 9.9 7.1 11.4 15.5 9.5 17.8 9.9

Asian 5.4 3.2 5.7 3.1 3.7 4.4 3.2 4.6

White 78.7 81.8 80.1 80.9 59.6 54.8 54 49.8

Other race 3.5 3 3.4 3.3 4.9 6.6 5.2 7.5

Lowest SES 11.4 16.8 10.4 18 25 15.6 28.9 15.9

Second SES 18.9 27.2 18.2 31.8 26.3 20 28.4 24

Third SES 28 30.3 28.1 30.6 26.2 27.9 25.1 25.4

Highest SES 41.6 25.7 43.3 19.6 22.5 36.5 17.6 34.6

Parent high school only 14.7 23.2 14 26.6 22.2 15.8 24.1 14.8

Parent some college 29.6 37.3 28.5 38.7 37.2 29.5 38.8 33.3

Parent BA 28.1 21.4 28.6 21.5 21.3 29.8 19.7 26.9

Parent MA or more 25.5 15 27.2 10.9 13.7 21.7 10.7 23.7

Northeast 20.8 12.1 19.8 11.9 19.3 28.1 19.5 23.5

South 30.4 31.5 30.3 30.7 35.4 37 36.1 43

Midwest 28 30.9 28.8 31.6 24 23.4 23.2 22.8

West 20.8 25.6 21.1 25.9 21.4 11.5 21.2 10.6

Rural high school 23.8 29.6 24 28.7 22.2 14.4 21.9 13.9

College athlete 10.1 6.2 10.2 3.8 7.4 17.3 7.1 21.6

AP class 27.9 27.1 29.7 26 15.1 17.6 12.8 11.7

NOTES: Data from ELS:2002. Sample of 2004 seniors. Sample weights used. Colleges divided into Barron's competitiveness categories. Respondents

counterfactually sorted into competitiveness category by a college graduation prediction based on high school GPA and test scores. Undermatching is when a

student is counterfactually sorted into a higher competitiveness category than the one she attends. Overmatching is when a student is counterfactually sorted

into a competitiveness category that is lower than the one she actually attends. Base rate refers to the average in the sample of respondents who could

possibly have been overmatched or undermatched. That is, only respondents counterfactually matched into the second-highest or lower competitiveness

categories could have overmatched. SES quartiles calculated based on 10th grade family income, mother's completed education and occupational prestige,

and father's completed education and occupational prestige.

3

Table 5. Percentages of Students in SES Quartiles by Type of

Initial Higher Education Experience, ELS:2002

Most or

Highly

Very Competitive Non- or

Less

2-yr Did not

attend

Highest SES 22.5% 23.9% 26.3% 5.8% 15.7% 5.8%

Third SES 7.9% 14.8% 25.9% 7.9% 26.4% 17.1%

Second SES 3.7% 8.0% 18.7% 8.0% 34.8% 26.7%

Lowest SES 2.5% 6.2% 13.0% 8.7% 30.4% 39.1%

Table 6. Revenues and Expenditures at Public Institutions

Most or Highly Very Competitive Non- or Less 2-yr

Revenues per student

State Appropriations 13-15 $9,714 $7,700 $6,281 $5,983 $3,651

Difference in State Appropriation 13-

15 minus 88-90

-$6,362 -$4,692 -$3,534 -$2,022 -$953

Expenditures per student

Instruction 13-15 $17,827 $12,783 $9,504 $7,925 $6,271

Difference in Instruction 13-15 minus

88-90

$ 5,682 $ 3,653 $2,365 $2,052 $1,608

Other Core Expenditures 13-15 $11,543 $8,300 $6,992 $6,531 $4,903

Difference in Other Core 13-15 minus

88-90

$ 4,907 $2,991 $2,654 $2,710 $2,750

FTE students 13-15 897,378 1,301,377 2,943672 446,208 3,011,752

Number of Institutions 34 75 272 80 691